davmlaw
commented
8 months ago Sent form out:
https://docs.google.com/forms/d/e/1FAIpQLSeQmQd3J5RVCHQ8xM35U426Xgc-mJUdoRVrUwKwkzrn4T4koA/viewform
will proceed with what I think is right but maybe change if people convince me.
Another option is to use known path/benign and try different settings, make ROC curves etc but that's a lot of work (Yasir to help?)
Maybe draw some diagrams to explain this. I will do the below, but be able to configure it, and we can change before SA Path deployment if people think so
OK - a question about structural variants (del/dup/inv) and gnomAD allele frequency of overlaps
a) I'm only going to report events of the same type (del for dels, dup for dups etc) b) I'm leaving the default minimum overlap percentage at 80% of input SV (includes over 100% so gnomAD one can completely cover input deletion for example) c) If there are multiple gnomAD entries, we'll keep them all to be able to display on the variant page, however we'll take the lowest frequency one and assign it to the single gnomAD af for the variant, the one that is used for filtering
Decision behind (c) is that it's best to uncover rare pathogenic variants (ie no false negatives) and the med scientists can
Another option might be to say, take the highest overlap.
If there are 2 overlapping SVs - one with 90% overlap and 0.001% AF and another that has 100% overlap and 1% AF - we'll set the input SV variant's AF to 0.001% (rarer one, even with smaller overlap)